In short
Ask one specific question about one meeting, ask for the moment or a quote the answer came from, and check it there. A good tool answers only from that meeting's transcript and says plainly when the transcript does not cover the question, rather than guessing.
To get useful answers from a meeting transcript, ask one specific question about one meeting, ask where in the meeting the answer came from, and check it there. "What did we decide about the launch date, and who said it?" works better than "what happened?". A good tool answers only from that meeting's transcript, and when the transcript does not cover the question it says so instead of guessing.
This guide covers what to ask, why answers should come from the meeting alone, and how to check what comes back.
Ask a question about a meeting
- Open the meeting you mean. Asking about one meeting keeps answers tied to one transcript you can check.
- Ask one thing at a time. "Who agreed to send the contract?" gets a checkable answer; "summarise the risks and the owners and the dates" gets a paragraph that is hard to verify.
- Ask for the evidence. Add "quote the line" or "give the time" to the question, so you know where to look.
- Check the answer at that moment. Find the quote in the transcript, or play the recording from that point. If you cannot find it, do not rely on the answer.
- Follow up or rephrase. If the answer is vague, ask a narrower question. If the tool says the transcript does not cover it, believe it and look elsewhere.
Questions that work
Specific questions with an answer that is either in the transcript or not:
- "What did we decide about the deadline?"
- "Who agreed to send the follow-up, and by when?"
- "What was left open at the end?"
- "What did Priya say about the budget?"
- "Did anyone disagree with the plan to move the launch?"
- "What numbers were mentioned for the annual price?"
- "What did the client ask us to do before the next call?"
Questions that tend to go wrong:
- Questions about tone or intent. "Was Tom annoyed?" The model reads text, not voices; it will guess from wording.
- Questions the meeting did not cover. "What is our Q3 revenue?" If it was not said, any number that comes back is invented or from somewhere else.
- Questions that need several meetings. "When did we first discuss this?" needs every meeting, and answers from one will be incomplete.
Why the answer should come from that meeting only
A general-purpose chatbot knows a great deal about how meetings usually go. That is the problem. Asked "what was the deadline?", it can produce a plausible date from the shape of the conversation, or from its own sense of what is typical, even when no date was said.
An answer tied to one transcript can be checked: it either appears in the text or it does not. An answer drawing on general knowledge, other documents or other meetings cannot be checked against the meeting you were in, and it reads exactly the same.
So when you use any tool for this, look for one that:
- answers from the meeting's transcript and nothing else;
- points to where the answer came from;
- says plainly when the transcript does not answer the question.
"The transcript does not say" is a good answer
It feels like a failure. It is the most useful thing a tool can tell you, because it is the answer that stops you acting on an invented one. It usually means one of:
- the subject came up in a different meeting;
- it was said too quietly, or over someone else, to be transcribed;
- it was never said, and your memory has filled it in.
In each case the next step is yours: search other meetings, listen to the part of the recording where you think it was said, or ask the person. Can you trust an AI meeting summary? covers this and the other ways AI answers go wrong.
Limits to keep in mind
- The transcript's mistakes pass through. If the recogniser heard "fifteen" as "fifty", the answer will say fifty. Numbers and names are worth checking against the audio, not only the text.
- Who said it depends on the transcript. If a line is credited to the wrong person, the answer will be too. Fix speaker names before asking about people.
- An answer is a reading. Like a summary, it is the model's reading of the transcript. Meeting transcript vs summary explains why the transcript is the thing to quote.
- It sends text. Asking a question sends the transcript text to whoever runs the model. How AI meeting notes are made explains where each step runs.
Questions by kind of meeting
| Meeting | Useful questions |
|---|---|
| Client call | "What did they ask us to do?" "What did they say about their deadline?" "What are they worried about?" |
| Planning | "What did we leave out, and why?" "Who owns each part?" "What risks were named?" |
| Interview | "What did the candidate say about their last role?" "Which of our questions did they not answer?" |
| One-to-one | "What did I promise to follow up on?" "What did they ask for help with?" |
| Standup | "Who is blocked, and on what?" |
Each of these has an answer that is either in the transcript or not, which is what makes it checkable.
Search or ask
Asking and searching do different jobs.
| You want | Use |
|---|---|
| The moment someone said a word or name | Search |
| Every meeting that mentioned a client | Search |
| Everything agreed about one topic in one meeting | Ask |
| A list of open questions | Ask, then check |
| An exact quote to send someone | Search, then copy from the transcript |
Building a searchable archive of your meetings covers search in depth.
Ask Notey
In Notey, Ask Notey sits beside the open meeting and answers questions about that meeting.
- The answer is drawn from that meeting's transcript. If the transcript does not cover the question, the answer says so plainly rather than guessing.
- Answers are labelled AI-generated.
- The conversation stays with the meeting. Come back later, even after quitting, and the questions and answers are still there in order. Clear them whenever you like.
- Text only. The question, the transcript text and the names you gave people are sent to Notey's service and on to OpenAI as a processor, not used for training and not retained by Notey. The recording stays on your Mac.
- Search is separate and local. ⌘K searches every meeting by time or by what was said, with no account and no network, and every transcript line plays from its timestamp, which is the quickest way to check an answer.
Ask Notey needs an account and a plan that includes AI notes; see pricing.
Frequently asked questions
Can AI answer questions about a meeting I recorded?
Yes. A language model reads the meeting's transcript and answers from it. The answer is only as good as the transcript, so a misheard name or number will come back in the answer. Check anything important against the transcript or the recording.
Why did the AI say the transcript does not answer my question?
Because nothing in that meeting's transcript supported an answer. That is the right behaviour. The topic may have come up in a different meeting, been said too quietly to transcribe, or never been said.
Should I search or ask?
Search when you remember the words ("pricing", a person's name) and want the exact moment. Ask when you want something put together, such as everything agreed about a deadline, and then check the answer at the moments it points to.
Can I ask questions across all my meetings at once?
Some tools offer that. It is harder to check, because the answer can mix meetings. In Notey, Ask Notey works on one meeting at a time, and search with ⌘K covers every meeting.
Does asking a question about a meeting send the recording?
It depends on the tool. In Notey, only the question, the transcript text and the names you gave people are sent; the recording stays on your Mac.